Browser ONNX Chat

Single-file dark UI for running your exported ONNX language model entirely in the browser.

Expected files: an ONNX model with input input_ids and output logits, plus a Hugging Face tokenizer tokenizer.json. Optional config can be loaded from tokenizer_config.json or entered manually below.

Model files

ONNX model Tokenizer JSON Tokenizer config JSON

Nothing loaded yet.

Chat template

User name Assistant name

Generation

Max tokens Temperature

Top-K Top-P

Repetition penalty

Context window Sampling mode

Special token ids

<|im_start|> <|im_end|>

These are used for stop detection and validation. If your tokenizer JSON includes these tokens, the page will auto-fill them when possible.

Chat

Load files, type a message, and the model replies locally in your browser.

Idle

Message